| paper | Spoiler Susceptibility in Party Elections |
| authors | Daria Boratyn, Wojciech Słomczyński, Dariusz Stolicki, Stanislaw Szufa |
| venue | ECAI 2025 |
| filed under | voting · theory |
| judged by | gpt-5.6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
Theorem 3 is a named but purely axiomatic characterization, so it fails the required computational-result gate. A distribution-over-types version of the paper’s \(L_1\) excess-impact calculation is statable and recognizable, but it is a direct weighted-profile evaluation rather than a computational theorem supplied by the paper. Therefore the mirror is red.
fails bit a — no named computational result to mirror
For fixed \(\mu\), \(K_i\), and \(f\), the proposed quantity follows by applying the paper’s existing redistribution and allocation rules; the supremum formulations merely restate its axiomatic properties.
fatal: True
The proposal covers the spoiler definition, Definition 8’s \(L_1\) metric, and Theorem 3’s property, but supplies no computational counterpart to Theorem 2 or any complexity result.
The strongest honest case is a qualified yes. The paper contains no named theorem giving an NP-hardness, polynomial-time, parameterized, or approximation result; its computational-complexity discussion is background, and its experiments are not numbered complexity results. So this cannot honestly be sold as a continuization of a complexity theorem. Its best anchor is instead a proved structural theorem whose continuous-population algorithmic counterpart is well posed.
My lead anchor is Theorem 3, proved in Appendix B. It states:
For any zero-sum voting method, a neutral, anonymous, weakly monotonic, and negative-unanimous allocation rule \(f\) is spoiler-proof if and only if it satisfies party independence of irrelevant alternatives.
Here “zero-sum” means that improving one party’s score or rank in a vote worsens another’s. Party independence of irrelevant alternatives means that, for a fixed party \(i\), \(f_i\) is unchanged whenever two profiles have the same rank or score for \(i\) vote by vote, even if the other parties are arranged differently.
The natural continuous mirror is Continuous Excess-Spoiler.
Take a finite set of parties \(P\), districts \(D\), and complete voter types \(T\). A type includes its district, complete ranking or score vector, and any behavioral information used by redistribution. A society is a mass vector \(\mu\in\Delta_T\), where \(\mu_t\) is the fraction of the electorate of type \(t\). For equal-sized districts, impose \(\sum_{t:\,d(t)=d}\mu_t=1/|D|\). Party elimination is represented by a redistribution kernel \(K_i(t,t')\), giving the fraction of type \(t\) converted into type \(t'\) after party \(i\) is removed. Thus
\[ \mu^{-i}_{t'}=\sum_{t\in T}\mu_tK_i(t,t'). \]
The allocation rule \(f\) is exactly the paper’s rule: compute district-level first-choice or Borda shares from the type masses, apply FPTP, Jefferson–D’Hondt, Webster–Sainte-Laguë, largest remainders, or another specified rule, and aggregate the resulting seat payoffs. Seats remain discrete within districts; only the voter population is continuous.
Given \(i\), \(\mu\), \(K_i\), and \(f\), compute
\[ \lambda_i^f(\mu,K_i) = \min_{\pi\in\Pi_{P,-i}} \left\|f(\mu^{-i})-\pi(f(\mu))\right\|_1. \]
The output is the exact excess impact, together with the decision whether it is positive. A positive value means that removing \(i\) decreases some other party’s payoff beyond the reallocation of \(i\)’s original payoff. The action is party elimination; the objective is the paper’s \(L_1\) excess-impact objective.
A theorem-level version asks whether
\[ \sup_{\mu,\;i,\;K_i}\lambda_i^f(\mu,K_i)=0. \]
Under the hypotheses of Theorem 3, this is equivalent to testing party independence of irrelevant alternatives. Equivalently, for ordinal types one can define \(\mu\sim_i\nu\) when the two mass distributions assign the same total mass to every rank position of \(i\), and ask whether
\[ \sup_{\mu\sim_i\nu} \left|f_i(\mu)-f_i(\nu)\right| \]
is zero. This is the exact continuous analogue of the paper’s finite-profile definition of party-equivalence.
This mirror is plausible because the paper already treats votes as exchangeable objects, applies redistribution independently to votes, and bases its allocation on aggregate profiles. The high-multiplicity regime is also realistic: millions of voters distributed across a fixed number of districts, with only \(d\cdot |P|!\) possible district-ranking types, perhaps multiplied by a modest number of behavioral response classes. The type is complete: voters with different districts, rankings, scores, or redistribution behavior are different types.
For a fixed finite type set and an explicitly represented allocation rule, I expect the fixed-instance problem to be Class A. Type masses and redistribution are linear, and the final excess-impact calculation is an LP. The universal maximization over all societies and redistribution kernels is more interesting: with an explicit piecewise-linear description of \(f\), it should reduce to finitely many LPs; with a succinct rule and variable numbers of parties, the complexity remains an open algorithmic question. Any hardness there would come from the rule’s winner-region geometry or the agenda, not from the number of individual voters.
The mirror deliberately covers Theorem 3 and the paper’s formal spoiler notion. It does not claim to continuize the paper’s experimental cultures directly. Those can become further questions: convergence rates from finite elections to mass profiles, approximate spoiler-proofness, and the complexity of maximizing expected excess impact over Mallows, spatial, or single-peaked population distributions.
The weakest point is that the paper itself already permits aggregate profiles and even mentions infinite profiles. In particular, the seats-votes setting is close to an existing continuous share model, so a single-district version could be accused of merely restating the paper. The stronger case is therefore the full type-level, district-aware version: the primitive object is a distribution over voter types, redistribution acts on that distribution, and the computational task is exact spoiler-impact analysis. Even then, because there is no named computational-complexity anchor, this is a promising structural mirror rather than a demonstrated continuization of a P-versus-NP result.
The negative case is strongest on the choice of anchor. This paper has no named computational theorem to continuize. Theorem 3 is an axiomatic characterization: spoiler-proofness is equivalent to party-independence of irrelevant alternatives. The paper never asks for an exact, approximation, parameterized, or worst-case algorithm.
The proposed “Continuous Excess-Spoiler” therefore does not really turn Theorem 3 into a ChoCo problem. For a fixed type distribution \(\mu\), redistribution kernel \(K_i\), and rule \(f\), one first computes \(f(\mu)\) and \(f(\mu^{-i})\), then evaluates the paper’s already-defined metric. In fact, the minimization over monotonic projections is essentially closed form. Writing \(s=f(\mu)\), \(s'=f(\mu^{-i})\), one obtains
\[ \lambda_i^f(\mu,K_i) = 2\sum_{j\neq i}\max\{s_j-s'_j,0\}. \]
Thus the alleged LP is not an optimization problem over population mass. It is a direct calculation after applying the electoral rule. With explicit types, forming \(\mu^{-i}=\mu K_i\), computing district vote shares, and applying FPTP, Jefferson–D’Hondt, Webster, or largest remainders are ordinary weighted-election computations. Replacing integer multiplicities by rational masses changes the representation and bit lengths, but not the computational task.
The proposed universal test,
\[ \sup_{\mu,i,K_i}\lambda_i^f(\mu,K_i)=0, \]
is simply spoiler-proofness restated over a measure space. Likewise, the condition
\[ \sup_{\mu\sim_i\nu}|f_i(\mu)-f_i(\nu)|=0 \]
is party-independence of irrelevant alternatives restated for distributions. The programme explicitly distinguishes computational continuization from continuizing axioms while leaving the society finite. This mirror lands on the axiomatic side of that boundary.
The district-aware refinement does not solve the problem. Including the district and complete ranking in the type is legitimate, and millions of voters over a fixed set of such types are a sensible high-multiplicity regime. But that is precisely why the construction collapses to a weighted finite profile. The paper’s seats-votes model already takes vote shares in \(\Delta_P\); with districts, it takes one such aggregate vector per district. A distribution over voter types merely regenerates those vectors. I am not treating existing high-multiplicity work as a collision—it supports the modelling premise—but here that premise does not expose a new computational phenomenon.
A stronger proposed mirror could choose \(\mu\) adversarially and maximize excess impact, perhaps under district marginals or a specified culture. That would be a coherent optimization problem, but it would add the central decision variable absent from the paper. The paper fixes an election profile and studies the effect of eliminating a party; it does not optimize over societies. If \(K_i\) is also optimized, the model permits arbitrary admissible vote redistribution and ceases to represent the paper’s empirical spoiler question. If \(K_i\) is fixed, the resulting task is robust optimization over the piecewise-defined geometry of the electoral rule, not a computational version of Theorem 3.
The same objection defeats Theorem 2 as a possible rescue. In the single-district seats-votes setting, the paper already works directly with continuous vote-share vectors \(x\in\Delta\), and its identity-function result is already stated on that aggregate domain. Introducing an underlying population measure adds no information relevant to the theorem. Moving to ordinal voter types and districts can make the model richer, but then the proposed problem is no longer the theorem’s seats-votes object.
The experimental susceptibility measure does not provide a stronger anchor. Replacing a random finite profile by a probability law over type masses is a valid population model, but it is a probabilistic or law-of-large-numbers reformulation of the paper’s simulations. It yields asymptotic or statistical questions, not the missing computational landscape.
There is one important limitation to this negative case: electoral populations plainly can have meaningful high multiplicity, and a newly invented problem that optimizes continuous voter mass could be worthwhile. I therefore cannot honestly prove that no scenario could ever produce a valuable ChoCo problem. The narrower and stronger conclusion is that this paper supplies no such problem itself: its main results are axiomatic, its spoiler metric is directly evaluable, and its aggregate share models already absorb the information that a population continuum would add.
The adversarial triple: the proponent anchors on up to three named results; the opponent sees that case and must defeat every anchor; the judge decides which case convinced it. These are the pipeline’s own outputs, generated by tools/triple_run.py — no human edited them. The paper’s own text is not reproduced here beyond the quoted statement above.